How to specify the Input shape for GRU model on keras functional api?

Question

I have a 2d matrix where the features are columns and the samples are rows (178136 features and 416 samples). I transposed this matrix and then converted this matrix to 3d by creating a NumPy array of 3 duplicate copies of this matrix using the following code.

matrix = matrix.transpose()
dataset = np.array([matrix,matrix,matrix])

The dimensions became the following: (3,178136,416) where 3 is the # of time steps for the sequential model, 178136 are the features, and 416 is the number of samples.

I am having trouble specifying the input shape for the Keras function api for the model I am trying to train. Below is the code I have written:

import numpy as np
import tensorflow
from tensorflow import keras
from tensorflow.keras import layers

inputdata = np.load('filepath')
outputdata = np.load('filepath')
# loads data from .npy files
input_size = 178136
# The input size is the number of genes in the dataset.
output_size = 155551
# The output size is the number of protein coding genes in the dataset.

inputs = keras.Input(shape=(3,input_size))
# This is the input layer of the neural network.

x = layers.GRU(input_size, return_sequences=True)(inputs)
# This is the first hidden layer of the GRU neural network.

x = layers.GRU(input_size, return_sequences=True)(x)
# This is the second hidden layer of the GRU neural network.

outputs = layers.Dense(output_size)(x)
# This is the output layer of the GRU neural network.

model = keras.Model(inputs=inputs, outputs=outputs, name="GRU_for_GRE_Modeling")
# This code specifies the input and output layers of the neural network.

model.compile(optimizer=keras.optimizers.Adam(), loss="categorical_crossentropy",metrics = ['accuracy'])
# This code specifies the optimizer, loss, and evaluation metrics for the neural network.
print('model compiled')
csv_logger = tensorflow.keras.callbacks.CSVLogger('evaluation_metrics.csv', separator=",", append=False)
history = model.fit(inputdata, outputdata, batch_size=1, epochs=1000000, callbacks=csv_logger, validation_split=.25)

# This code actually trains the neural network and saves the evaluation metrics.

model.save('filepath')

# This code saves the trained neural network to a specified path.

Here is the error I keep receiving when I try to run this code:

    history = model.fit(inputdata, outputdata, batch_size=1, epochs=1000000, callbacks=csv_logger, validation_split=.25)
  File "/data/Libs/tensorflow/1.12.2/lib/python3.6/site-packages/tensorflow/python/keras/engine/training.py", line 1536, in fit
    validation_split=validation_split)
  File "/data/Libs/tensorflow/1.12.2/lib/python3.6/site-packages/tensorflow/python/keras/engine/training.py", line 992, in _standardize_user_data
    class_weight, batch_size)
  File "/data/Libs/tensorflow/1.12.2/lib/python3.6/site-packages/tensorflow/python/keras/engine/training.py", line 1117, in _standardize_weights
    exception_prefix='input')
  File "/data/Libs/tensorflow/1.12.2/lib/python3.6/site-packages/tensorflow/python/keras/engine/training_utils.py", line 332, in standardize_input_data
    ' but got array with shape ' + str(data_shape))
ValueError: Error when checking input: expected input_1 to have shape (3, 178136) but got array with shape (178136, 416)

Can someone help me resolve this issue? I have gone through all the documentation and I can't seem to find how to specify the number of samples to make this error go away

Answer 1

I tried to replicate your issue but its working as expected

import numpy as np
import tensorflow
from tensorflow import keras
from tensorflow.keras import layers


#inputdata = np.load('filepath')
#outputdata = np.load('filepath')
# loads data from .npy files
input_size = 100
# The input size is the number of genes in the dataset.
output_size = 80
# The output size is the number of protein coding genes in the dataset.


    inputs = keras.Input(shape=(3,input_size))
    # This is the input layer of the neural network.
    
    x = layers.GRU(input_size, return_sequences=True)(inputs)
    
    x = layers.GRU(input_size, return_sequences=True)(x)
    
    
    outputs = layers.Dense(output_size)(x)
    # This is the output layer of the GRU neural network.
    
    model = keras.Model(inputs=inputs, outputs=outputs, name="GRU_for_GRE_Modeling")
    # This code specifies the input and output layers of the neural network.
    
    model.compile(optimizer=keras.optimizers.Adam(), loss="categorical_crossentropy",metrics = ['accuracy'])
    # This code specifies the optimizer, loss, and evaluation metrics for the neural network.
    print('model compiled')
    model.summary()
    #csv_logger = tensorflow.keras.callbacks.CSVLogger('evaluation_metrics.csv', separator=",", append=False)
    #history = model.fit(inputdata, outputdata, batch_size=1, epochs=1000000, callbacks=csv_logger, validation_split=.25)

Output: model.summary shows correct input shape

model compiled
Model: "GRU_for_GRE_Modeling"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
input_7 (InputLayer)         [(None, 3, 100)]          0         
_________________________________________________________________
gru_10 (GRU)                 (None, 3, 100)            60600     
_________________________________________________________________
gru_11 (GRU)                 (None, 3, 100)            60600     
_________________________________________________________________
dense_2 (Dense)              (None, 3, 80)             8080      
=================================================================
Total params: 129,280
Trainable params: 129,280
Non-trainable params: 0